4.2 Our approach
4.2.1 Defining interaction indicators
The e-NOTEBOOK system is an effective interactive tool that promotes learners to collaborate with and learn during interactions and to create more knowledge construction. [Moore, 1992] classified interactions to three types in distance education. His three-part interaction scheme included: (1) learner-instructor, (2) learner-learner, and (3) learner-content interaction. Learner-instructor interactions establish an environment that encourages learners to understand the content better. And it is “regarded as essential by many educators and highly desirable by many learners” [Moore, 1989].
Learner-learner interaction takes place “between one learner and other learners, alone or in group settings, with or without the real-time presence of an instructor” [Moore, 1989, p.4]. Many studies show that this type of interaction is a valuable experience and learning resource [Bull, 1998]
[Vrasidas and McIssac, 1999]. Empirical evidence shows that students actually desire learner-learner interactions, regardless of the delivery method [Su et al., 2005]. Learner-content interaction is
defined as “the process of intellectually interacting with content that results in changes in the learner’s understanding, the learner’s perspective, or the cognitive structures of the learner’s mind”
[Moore, 1989, p.2]. Since our system is learner-centered, the main focus here is on the interactions that include: (1) learner-learner; (2) learner-content. The factors of interaction frequently observed in our system are as follows (Tab.4.1).
The various interactions imply different learner roles. The interaction type definition naturally various according to its usage. In our system, we create a set of interactions that includes three interaction indicators: (1) Comprehension of Web page; (2) Adequacy of remark; (3) Agreement of comment.
60 Chapter 4: Recommendation of Advanced Learners
Moreover, [Gunawardena et al., 1997] proposed an Interaction Analysis Model (IAM) which divides knowledge construction from discussion into five aspects to conveniently analyze the depth of social knowledge construction. The Interaction Analysis Model specifies that knowledge construction process proceeds through five phase:
• Sharing/comparing of information: Statement of observation or opinion; statement of agreement between participants
• Discovering/exploring of dissonance or inconsistency among ideas, concepts or statements:
Identifying areas of disagreement, asking and answering questions to clarify disagreement
• Negotiation of meaning and co-constructing of knowledge: Negotiating meaning of terms and negotiation of the relative weight to be used for various agreement
• Testing and modifying of proposed synthesis or co-construction: Testing the proposed new knowledge against existing cognitive schema, personal experience or other sources
• Agreeing on final statement and applications of newly constructed meaning or insights:
Summarizing agreement and meta cognitive statements that show new knowledge construction
This Interaction Analysis Model suggests a process that might contribute to community building as well. Since this coding scheme has been used in many past studies, this improves research validity in a quantitative content analysis [Rourke and Anderson, 2004]. We also used this coding scheme in our study to analyze the knowledge construction type and depth in learners. [Gunawardena et al., 1997][Schellens and Valcke, 2005] reported that knowledge construction in a group of online learning, the amount of posting apparently constituted the first two phase of IAM. The first phase of sharing and comparing - and it links to the idea of shared beliefs and also to Vygotsky's zone of proximal development [Vygotsky, 1978]. In this research, we mainly consider the first two phases:
(1) Sharing and comparing; and (2) Discovering and exploring.
Analysis of the information collected in such e-NOTEBOOK requires a number of advanced data processing steps including extraction of dominant, recurring themes and ultimately characterization of the degree of innovation represented by the various notes (abstracts and remarks) and comments
Chapter 4: Recommendation of Advanced Learners 61
created in our system. Describes of 3 interaction indicators are in the following.
1) Comprehension of Web page (CoW)
Definition 4.1: Comprehension of web page is learner’s understanding level based on similarity between web page read by learner and abstract wrote by learner.
The interaction indicator – comprehension of web page belongs to learner-content interaction. We calculate a similarity to measure the overlap between the two keyword vectors, keyword vector of web page and keyword vector of abstract, which in the end leads to mappings between them, i.e., the higher comprehension degree, the higher similarity and the higher probability they should be mapped. There are many successful similarity measures: COSINE, Pearson, and Jaccard similarity etc. We finally retained the Jaccard similarity because it performed slightly better overall and it is good at computing the similarity between the sets (formula 4.1).
j j i
j i
Webp Webp Learner
Webp Learner Jaccard
ij K
K K K
K S
CoW Ç
=
= ( , ) (4.1)
2) Adequacy of remark (AoR)
Definition 4.2: Adequacy of remark is remark’s usefulness based on positive/negative sentiment of comments. Assumption: each comment focuses on a single remark and contains opinion from a single opinion holder.
The interaction indicator – adequacy of remark belongs to learner-learner interaction. We adopt the first two phases of Interaction Analysis Model (IAM) here. Therefore the important task here is to extract the opinion of each comment and to classify comments as positive and negative opinion.
The adequacy of remark is computed as follows.
Step1: Classify the comment as positive or negative.
62 Chapter 4: Recommendation of Advanced Learners
WWW
NOTE Abstract Remark
Comments NOTE Abstract Remark
NOTE Abstract Remark
……
……
Web Page
Fig.4.2 Comments to the remarks
Since the adequacy of remark is the positive ratio of comments which point to the remark directly, that is to say, the comments of first layer of the tree are valid.
Many researches of automated opinion detection have been proposed [Wiebe et al., 2002] [Bruce and Wiebe, 1999] [Yu and Hatzivassiloglou, 2003] [Pang et al., 2002]. Turney presents a simple unsupervised learning algorithm for classifying a review as recommended or not recommended [Turney, 2002]. They use Pointwise Mutual Information (PMI) and Information Retrieval (IR) to measure the similarity of pairs of words of phase. The semantic orientation of a given phrase is calculated by comparing its similarity to a positive reference word (such as “excellent”) with its similarity to a negative reference word (such as “poor”). More specifically, a phrase is assigned a numerical rating by taking the mutual information between the given phrase and the word
“excellent” and subtracting the mutual information between the given phrase and the word “poor”.
In addition to determining the direction of the phrase’s semantic orientation (positive or negative, based on the sign of the rating), this numerical rating also indicates the strength of the semantic orientation based on the magnitude of the number. Here we adopt the unsupervised learning algorithm proposed by [Turney, 2002] (Appendix C) to classify the comment as positive or negative.
Chapter 4: Recommendation of Advanced Learners 63
In our system, the first step is to use a chasen- a Japanese morphological analysis system – to identify phrases in the comment that contain adjectives or adverbs or adjectival noun. The second step is to estimate the semantic orientation of each extracted phrase. The third step is to adding the given comment to a class, positive or negative. Where, in our system,
“excellent”= “いい( good)｜好き( like)｜良い(good)｜魅力(charm)｜大好き(favorite)｜欲し い(want)｜楽しい(happy)｜嬉しい(happy)｜面白い(interesting)｜素敵(lovely)｜良かった (good)｜良く(well)｜満足(satisfaction)｜素晴らしい(wonderful)｜よい( good) ”
“poor”= “あまり(not good)｜あんまり(not good)｜悪い(bad)｜嫌い(dislike)｜嫌(dislike)｜不 安(worry)｜怖い(fearful)｜不良(bad)｜苦手(hard)｜悪く(bad)｜まずい(bad)｜悲しい(sad)｜
危険(risk)｜だめ(useless)｜辛い(painful)｜不満(dissatisfaction)｜不快(dissatisfaction)｜最悪 (worst)”
Step2: Calculate the adequacy of remark using formula (4.2).
100%
#
) (
# ´
= Comment
positive Comment
AoR (4.2) Where # denotes the number of.
3) Agreement of comment (AoC)
Since explicit structure has the advantages that they encourage participants to clarify their thinking [Brna et al., 2001], our comment space is a tree structure with a clear flow of messages from the initial parent post to subsequent ideas and thoughts. In terms of defining and evaluating the depth of knowledge interactions in a note, the extent of knowledge construction in comment space (i.e. posted comments and replies) is a key indicator for us to understand the depth of learners’ interactions.
Theories and paradigms for knowledge construction have been discussed for decades [Allchin, 1999].
Our research focuses on learner-oriented knowledge construction and learning behavior through knowledge sharing under social influence. In the same time, we also consider the gradual construction of knowledge through cognitive conflict and the forming of consensus from social
64 Chapter 4: Recommendation of Advanced Learners
interactions. Our studies support the concept that learners focus on constructing knowledge via writing about their own experience and sharing replies.
Definition 4.3: Agreement of comment is comment’s usefulness based on the comments’
sentiment and relations. Assumption: each comment focuses on its parent comment and contains opinion from a single opinion holder.
The interaction indicator – agreement of comment belongs to learner-learner interaction. The agreement and value of comment can be computed as follows.
Step1: Classify the comment as positive or negative.
We also adopt the algorithm (Appendix C) proposed by [Turney, 2002] to classify the comment as positive or negative.
Step2: Identify exceptional utterances.
[Yamaguchi et al., 2009] have proposed a method for analyzing the utterances and mining potential of ignored utterances through a process of interactions between humans. And they also emphasis that the exceptional utterances are always have high values. In our system, the decision of the comment’s role is based on the method proposed by [Yamaguchi et al., 2009]. Since Japanese words are written without a space, we divided comment into each word by Chasen. The word frequency of input is calculated. In our system, we use the approach (Appendix D) proposed by [Yamaguchi et al., 2009] to calculate the E . If E>>a, the comment is an exceptional utterance. Here a=0.1.
Step3: Calculate the value of each comment.
[Helander et al., 2007] discussed the Innovation Jam with the objective of identifying innovative and promising “Big Ideas” through a moderated on-line discussion between IBM worldwide employees and external contributors. [Murakami and Takeda, 2007] gave a micro analysis of how
Chapter 4: Recommendation of Advanced Learners 65
to identify the important utterances. Thread information is shown to be an important resource for analysis of interaction among participants of a group. Our comment-taking method provides a tree structure for participants to submit comment. By using this tree structure of comments, the value of each comment itself can be calculated (Tab.4.2).
The final value of each comment is calculated as follows: Firstly, extracting the expression of each comment, and identifying the relationship such as positive or negative between parent comment and child comment (in our system, only positive (agreement) and negative (disagreement) two situations are be considered) based on Step1. Secondly, the value of comment itself is calculated based on table 4.2. Thirdly, calculate the final value of each comment by using a recursive algorithm. If the comment is agreed by many users, it is important and on the contrary it is not important. As Fig.4.3 shows, the importance is calculated using recursive algorithm (formula 4.3).
= + ´
å
´j
ij j i
i S C R
C a (4.3)
Where Ci is the final value of the comment. Si is the value of comment itself. a is the propagate value. In our case, the adequacy of remark is defined as a. Rij is the relation of comments. For example, Rij=1(agree) and Rij=-1(disagree).
Tab.4.2 The criterion of calculating value of comment
Type Role
Value of Child Node Itself Value of Parent Node Disagree E>>a high low
E<<a low low
Agree E>>a high high
E<<a low high
high=1 low=0.1
66 Chapter 4: Recommendation of Advanced Learners
1
2 3
4 5
9
6 7 8
S9=1 C9=S9=1
S6=1 C6=S6=1
S7=1 C7=S7=1
S8=2 C8=S8=2 S3=1
C3=4.2 Propagate Value= Adequacy of Remark=0.8
Example:
1
2 3
4 5
9
6 7 8
S9=1 C9=S9=1
S6=1 C6=S6=1
S7=1 C7=S7=1
S8=2 C8=S8=2 S3=1
C3=4.2 Propagate Value= Adequacy of Remark=0.8
Example:
Fig.4.3 An example of calculating the value of comment
Our approach makes use of the thread information of the collaboration session to construct a tree graph that represents the flow of interaction, with each node denoting the content of each comment. C is the final value of the comment. S is the value of comment itself. R is the relation of comment.
mysql> select keyword1,keyword2, rs from keyword_rs where id=1;
+---+---+---+
| keyword1 | keyword2 | max(rs) | +---+---+---+
| システム| 工程 | 2.86272764205933 |
| 家庭 | こと | 0.260667532682419 |
| 家庭 | 改善 | 0.102305047214031 |
| 家庭 | 汚染 | 0.954242527484894 |
| 家庭 | 環境 | 1.46478748321533 |
| 家庭 | 海洋 | 1.65860760211945 |
| 家庭 | 危機 | 2.86272764205933 |
| 家庭 | 開発 | 2.86272764205933 |
| 家庭 | 交通 | 0.862727522850037 |
| 家庭 | 計画 | 0.779942154884338 |
| 家庭 | 事態 | 0.177882164716721 |
| 家庭 | 自転車| 0.401829689741135 |
| 家庭 | 重大 | 1.17253148555756 |
| 家庭 | 都市 | 1.3064249753952 |
……
| 自転車| 長期 | 2.86272764205933 |
| 自転車| 発展 | 1.90848505496979 |
| 地球 | システム| 0.779942154884338 |
| 地球 | 家庭 | 1.3064249753952 |
| 地球 | 完全 | 1.05654752254486 |
| 地球 | 記録 | 0.032780833542347 |
| 地球 | 共同 | 1.65860760211945 |
| 地球 | 最高 | 0.102305047214031 |
| 地球 | 自転車| 1.17253148555756 |
| 地球 | 人類 | 2.26066756248474 |
| 地球 | 世界 | 2.86272764205933 |
| 地球 | 生態 | 0.862727522850037 |
| 地球 | 製造 | 0.954242527484894 |
| 地球 | 地球 | 2.86272764205933 |
| 地球 | 百万 | 0.177882164716721 | +---+---+---+
52 rows in set (0.01 sec) mysql> select * from
keywordlist_all where id=1;
+---+---+---+
| id | keyword | times | +---+---+---+
| 1 | 改善 | 1 |
| 1 | 汚染 | 2 |
| 1 | 危機 | 1 |
| 1 | 緩和 | 1 |
| 1 | 交通 | 2 |
| 1 | 計画 | 2 |
| 1 | 最高 | 1 |
| 1 | 事態 | 1 |
| 1 | 自転車| 3 |
……
| 1 | 渋滞 | 1 |
| 1 | 世界 | 1 |
| 1 | 製造 | 2 |
| 1 | 代わり| 1 |
| 1 | 大気 | 1 |
| 1 | 騒音 | 1 |
| 1 | 低減 | 1 |
| 1 | 都市 | 3 |
| 1 | 氾濫 | 1 |
| 1 | 百万 | 1 | +---+---+---+
22 rows in set (0.53 sec)
Natural language process
& data mining
Estimating interaction indicators
•Comprehension of Web page
•Adequacy of remark
•Agreement of comment Knowledge production
& knowledge construction
Ex:
CoW11=0.4524 AoR11=0.7500 C3=4.2 NOTE
Abstract Remark
comment Web page
Fig.4.4 Example of estimating interaction indicators
The content of Web pages/ Notes (Abstract and Remark) / Comment is analyzed by using natural language process and data mining. Three interaction indicators are defined and calculated: Comprehension of Web page (CoW) is learner’s understanding level based on similarity between web page read by learner and abstract wrote by learner; Adequacy of remark (AoR) is remark’s usefulness based on positive/negative sentiment of comments; Agreement of comment (AoC) is comment’s usefulness based on the comments’ sentiment and relations.